r600g: build fetch shader from vertex elements

Vertex elements change are less frequent than draw call, those to
avoid rebuilding fetch shader to often build the fetch shader along
vertex elements. This also allow to move vertex buffer setup out
of draw path and make update to it less frequent.

Shader update can still be improved to only update SPI regs (based
on some rasterizer state like flat shading or point sprite ...).

Signed-off-by: Jerome Glisse <[email protected]>

1 files changed, 83 insertions, 0 deletions

diff --git a/src/gallium/drivers/r600/r600_state.c b/src/gallium/drivers/r600/r600_state.c
index c592ef2bd05..9b70942eebf 100644
--- a/src/gallium/drivers/r600/r600_state.c
+++ b/src/gallium/drivers/r600/r600_state.c
@@ -94,6 +94,84 @@ void r600_polygon_offset_update(struct r600_pipe_context *rctx)
 	}
 }
 
+void r600_vertex_buffer_update(struct r600_pipe_context *rctx)
+{
+	struct r600_pipe_state *rstate;
+	struct r600_resource *rbuffer;
+	struct pipe_vertex_buffer *vertex_buffer;
+	unsigned i, offset;
+
+	/* we don't update until we know vertex elements */
+	if (rctx->vertex_elements == NULL || !rctx->nvertex_buffer)
+		return;
+
+	/* delete previous translated vertex elements */
+	if (rctx->tran.new_velems) {
+		r600_end_vertex_translate(rctx);
+	}
+
+	if (rctx->vertex_elements->incompatible_layout) {
+		/* translate rebind new vertex elements so
+		 * return once translated
+		 */
+		r600_begin_vertex_translate(rctx);
+		return;
+	}
+
+	if (rctx->any_user_vbs) {
+		r600_upload_user_buffers(rctx);
+		rctx->any_user_vbs = FALSE;
+	}
+
+	if (rctx->vertex_elements->vbuffer_need_offset) {
+		/* one resource per vertex elements */
+		rctx->nvs_resource = rctx->vertex_elements->count;
+	} else {
+		/* bind vertex buffer once */
+		rctx->nvs_resource = rctx->nvertex_buffer;
+	}
+
+	for (i = 0 ; i < rctx->nvs_resource; i++) {
+		rstate = &rctx->vs_resource[i];
+		rstate->id = R600_PIPE_STATE_RESOURCE;
+		rstate->nregs = 0;
+
+		if (rctx->vertex_elements->vbuffer_need_offset) {
+			/* one resource per vertex elements */
+			unsigned vbuffer_index;
+			vbuffer_index = rctx->vertex_elements->elements[i].vertex_buffer_index;
+			vertex_buffer = &rctx->vertex_buffer[vbuffer_index];
+			rbuffer = (struct r600_resource*)vertex_buffer->buffer;
+			offset = rctx->vertex_elements->vbuffer_offset[i] +
+				vertex_buffer->buffer_offset +
+				r600_bo_offset(rbuffer->bo);
+		} else {
+			/* bind vertex buffer once */
+			vertex_buffer = &rctx->vertex_buffer[i];
+			rbuffer = (struct r600_resource*)vertex_buffer->buffer;
+			offset = vertex_buffer->buffer_offset +
+				r600_bo_offset(rbuffer->bo);
+		}
+
+		r600_pipe_state_add_reg(rstate, R_038000_RESOURCE0_WORD0,
+					offset, 0xFFFFFFFF, rbuffer->bo);
+		r600_pipe_state_add_reg(rstate, R_038004_RESOURCE0_WORD1,
+					rbuffer->size - offset - 1, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_038008_RESOURCE0_WORD2,
+					S_038008_STRIDE(vertex_buffer->stride),
+					0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_03800C_RESOURCE0_WORD3,
+					0x00000000, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_038010_RESOURCE0_WORD4,
+					0x00000000, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_038014_RESOURCE0_WORD5,
+					0x00000000, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_038018_RESOURCE0_WORD6,
+					0xC0000000, 0xFFFFFFFF, NULL);
+		r600_context_pipe_state_set_fs_resource(&rctx->ctx, rstate, i);
+	}
+}
+
 static void r600_draw_common(struct r600_drawl *draw)
 {
 	struct r600_pipe_context *rctx = (struct r600_pipe_context *)draw->ctx;
@@ -132,6 +210,7 @@ static void r600_draw_common(struct r600_drawl *draw)
 	if (r600_pipe_shader_update(&rctx->context, rctx->ps_shader))
 		return;
 
+#if 0
 	for (i = 0 ; i < rctx->vertex_elements->count; i++) {
 		uint32_t word2, format;
 
@@ -159,6 +238,7 @@ static void r600_draw_common(struct r600_drawl *draw)
 		r600_pipe_state_add_reg(rstate, R_038018_RESOURCE0_WORD6, 0xC0000000, 0xFFFFFFFF, NULL);
 		r600_context_pipe_state_set_fs_resource(&rctx->ctx, rstate, i);
 	}
+#endif
 
 	mask = 0;
 	for (int i = 0; i < rctx->framebuffer.nr_cbufs; i++) {
@@ -195,6 +275,7 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
 	struct r600_drawl draw;
 	boolean translate = FALSE;
 
+#if 0
 	if (rctx->vertex_elements->incompatible_layout) {
 		r600_begin_vertex_translate(rctx);
 		translate = TRUE;
@@ -204,6 +285,8 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
 		r600_upload_user_buffers(rctx);
 		rctx->any_user_vbs = FALSE;
 	}
+#endif
+
 	memset(&draw, 0, sizeof(struct r600_drawl));
 	draw.ctx = ctx;
 	draw.mode = info->mode;